Project Application Details

Euan Macdonald dsb14145

Project aim is to simulate a series of extreme wind-wave events in the Firth of Clyde to provide additional training data for an adaptive Bayesian model selection process. This process concerns averaging network predictions based on the range of posterior probabilities for all networks in the group to make robust predictions with an associated confidence interval. This method has already been successfully deployed to make 3, 6 and 12 hour prediction for all cases apart from the most extreme, trained solely on historical data. The poor performance of the extreme region is attributed to the lack of training data available in the historical dataset, thus presenting the requirement for a database of synthetic extreme events to supplement the training data and improve extreme value predictions. This model has a real time forecasting application for predicting overtopping events.